On Wednesday Pat Quinn gave a pretty interesting interview. The first half was the same old resigned Pat Quinn that we've come to know over these last few weeks. He spoke about some individual players, talked about what the team needed to do to improve and generally sounded depressed. Then he started talking about the Chicago Blackhawks. He began by correcting the notion that Chicago is a "puck possession" team (my thoughts on this are summed up better than I could do it by "E" in the main post and "RO" in the comments here). It was interesting and entertaining to hear Quinn speak about dump-and-chase hockey in the same way football people talk about establishing the run. As he went on about hockey strategy he started to get excited and eventually made a comment about what he would do if the Oilers meet the Blackhawks in the playoffs. I like to think that for at least one brief moment Quinn forgot just how awful his team was. Welcome to Fantasyland.
And that's what the playoffs have become for the Oilers and their fans: a fantasy. But I think I'm going to indulge and talk a little bit about the playoffs and, more specifically, about how standings points predict playoff success. If that sounds familiar, it's likely because this is an expansion and/or response to some of the work I did on this last week. We'll look at some data after the jump.
In my last post I mentioned that a team's points percentage over their last 32 regular season games was a better predictor of postseason success than a team's points percentage over the whole season and a substantially better predictor than the team's first 50 regular season games. At the time, I was curious about how other samples might fare so I looked at a team's performance over their final 5, 10, 20, 30, 40, 50, 60, 70 and 82 regular season games and compared that to playoff series wins. If you just bet on the team with the most points over each time frame, how many would you get right?
Ties are the instances where both teams had an equal number of points. In the 5, 10 and 20 game samples I would suggest that the sample size is not large enough. The 30 and 40 game samples, on the other hand, do just as well as the larger samples of games and the 30 game sample actually does the best job of predicting series wins. How much of this is just normal variance and how much is a real effect? I'm really not sure. I do know that if I chose to limit my sample size (since at some point, the past is the past) in trying to predict playoff winners I would narrow it down to something in the 30-40 games range. Since points probably aren't the best indicator of future success, I think next I'll try to compare the predictive value of goal differential or shot differential over various time frames.